Sputtering, alternatively called physical vapor deposition (PVD), is commonly used in the fabrication of semiconductor integrated circuits for depositing layers of metals and related materials particularly for the formation of electrical interconnections. Somewhat older integrated circuit technology uses aluminum for horizontal interconnects and for vertical interconnects between metallization levels through vias having relatively modest aspect ratios. Such applications require fast deposition rates and high uniformity that are easily achievable with sputtering. The fast deposition rate has been achieved in part by magnetron plasma sputtering in which a working gas, for example, of argon is excited into a plasma. The positively charged ions are attracted to a negatively biased metallic target and strike it with sufficient energy to dislodge (sputter) metal atoms from the target, which then coat a wafer positioned in opposition to the target. The sputtering rate is enhanced by positioning a magnet assembly in back of the target which creates a magnetic field parallel to the front face of the target. The magnetic field traps electrons, which increases the plasma density and hence the sputtering rate. The most prevalent type of magnetron in commercial fabrication uses a series of horseshoe or similar magnets having closely spaced poles. The magnets are arranged in a closed kidney-shaped path. Although the total area of the magnetron is fairly large, the magnetic field extends over a relatively small area. To achieve the required uniformity of deposition, the kidney-shaped magnetron is rotated about the center of the target.
More advanced integrated circuit technology has placed somewhat different and more difficult requirements upon sputtering, and emphasis in sputtering has shifted from depositing horizontal interconnects to depositing vertical vias. The high complexity of advanced integrated circuits has been achieved in large part by decreasing minimum feature size and spacing between features. The resulting complex wiring has been accomplished by interconnecting multiple wiring levels by vias extending through an intervening dielectric layer. As schematically illustrated in FIG. 1, a lower dielectric layer 10, typically formed of silicon dioxide or related silicate glasses, includes a conductive feature 12 at its surface. An upper dielectric layer 14 is deposited over it. A via hole 16 is etched through the upper dielectric layer 14 overlying the conductive feature. The width of the via hole 16 is being pushed to 0.18 μm and below. Minimum feature sizes of 0.10 μm and even 0.07 μm are being developed. However, the thickness of the inter-layer dielectric layer 14 is constrained to be a minimum of 0.7 to 1.0 μm to minimize cross talk and prevent dielectric discharge. The result is that the via holes 16 may have aspect ratios of 5:1 and greater. Sputtering is fundamentally a generally isotropic ballistic process ill suited to reaching into high aspect-ratio holes. If sputtering is used to fill the hole 16 with metal, the sputtering is likely to preferentially coat the upper corners of the hole 16 and to close it before the bottom is filled.
Furthermore, with such small feature sizes, diffusion between the metal and dielectric portions must be minimized. Accordingly, a standard practice has developed to precoat the via hole 16 as well as the planar top of the upper dielectric layer 14 with a thin barrier layer 20. A typical barrier material for aluminum metallization is Ti/TiN and that for copper metallization is Ta/TaN although other barrier materials and combinations have been proposed. To achieve its purpose, the barrier layer 20 should significantly and fairly uniformly coat the sides and probably also the bottom of the via hole. Again, sputtering is not inherently adapted for sidewall coverage.
Much work has been recently expended in developing the technology for copper metallization. Copper offers advantages of lower conductivity and reduced electromigration. Further, copper can easily be deposited even into high aspect-ratio holes by electrochemical plating (ECP). However, electrochemically plated copper requires that a copper seed layer 30, as illustrated in FIG. 2, be coated onto the top of the dielectric 14 and the sidewalls and bottom of the via hole 16 before a thick copper layer 32 is deposited by ECP. The copper seed layer 30 requires good bottom and sidewall coverage. Copper sputtering is preferred even under these difficult conditions.
The thick ECP copper layer 32 acts as both the via and the horizontal interconnects, typically in a process called dual damascene in which a trench is formed in the upper part of the dielectric layer 14 interconnecting multiple vias in the bottom part of the dielectric layer 14. The portion of the thick ECP copper layer 32 extending above the trench and the top of the dielectric layer is removed by chemical mechanical polishing (CMP). As a result, sputtering is being used less for depositing thick conductive layers and more for depositing thin layers in unfavorable geometries, in what are called barrier applications.
Both the barrier layer 20 and copper seed layer 30, when deposited by sputtering, tend to suffer the same type of non-uniform deposition typified by a sputtered layer 36 in the cross-sectional view of FIG. 3. A blanket portion 38 on the top of the dielectric layer 14 is relatively thick compared to a sidewall portion 40 and a bottom portion 42. The sidewall portion is high-aspect ratio holes 16 typically exhibits the lowest coverage relative to the blanket portion 38 and further often suffers from a minimum thickness 44, which needs to be maintained above a critical level to provide an electroplating current path to the bottom of the hole 16. Furthermore, an overhang portion 46 tends to form at the top of the hole 16 with a reduced entrance aperture 48. Although electroplating is generally effective at filling copper into a high aspect-ratio hole 16, it tends to be nearly conformal so that the entrance aperture 48 may close prior to completing the filling of the bottom of the hole 16. The resulting void in the copper severely affects the performance and reliability of the resulting device. An overly thin sidewall area 44 also results in a void included in the copper.
It has been recognized that effective sputtering of the barrier and copper seed layers can be accomplished by assuring a high fraction of ionized sputter metal atoms, whether of the barrier metal or of the copper, and by RF biasing the pedestal electrode supporting the wafer. The RF bias creates a negative self-bias adjacent to the plasma and accelerates the metal ions toward the wafer. The high forward velocity promotes the penetration of the metal ions deep into the high aspect-ratio holes.
A high density plasma of the sputter working gas increases the metal ionization fraction. Some suggestions have been made to achieve the high density plasma by inductively coupling additional RF power into the chamber. However, inductively coupled reactors tend to require high argon pressures and result in a high-temperature operation with possible damage from the energetic argon ions being accelerated to the wafer. The metal ionization fraction can also be increased by increasing the DC target power. However, for the 300 mm wafer technology being developed and even for 200 mm wafers, this approach causes the required power supplies to become prohibitively expensive, and controlling the target temperature becomes difficult.
Another and preferred approach, some times called self-ionized plasma (SIP) sputtering, described by Fu in U.S. Pat. No. 6,183,614, incorporated herein by reference in its entirety, is particularly useful for barrier sputtering in which only very thin layers are deposited, for example, 10 nm or less. SIP sputtering may be implemented with conventional planar targets in generally conventional and inexpensive magnetron sputter reactor chambers. In contrast, inductively coupled reactors require inductive coils in an expensive new design, and hollow cathode or vaulted target reactors expensive complexly shaped targets. SIP sputtering is based upon a small but strong magnetron which concentrates the high-density plasma region over a relatively small area of the target. As a result, somewhat modest power supplies of about 20 to 40 kW can be used to create a very high effective power density in the portion of the target underlying the magnetron. The high density plasma creates a high ionization fraction of the metal ions, estimated to be about 20% or greater. The metal ions are attracted to the wafer by RF biasing of the pedestal electrode to promote the coating the sides of deep holes.
Furthermore, the metal ion density is so high that some of the metal ions are attracted back to the target to resputter the target, hence the term self-ionized plasma. As a result, once the plasma has been ignited, the argon pressure in the chamber can be reduced, thus reducing the probability of scattering of the metal ions on their way to the wafer. A collision of a metal ion and argon would likely neutralize the metal atom. In the case of copper sputtering, under the right circumstances, the argon can be removed completely in a process called self-sustained sputtering (SSS).
SIP sputtering also benefits from an unbalanced magnetron including an inner pole of one vertical magnetic polarity surrounded by an outer pole of the opposed polarity, the total magnetic strength of the outer pole, that is, the magnetic flux integrated over the area of the outer pole, is substantially greater than that of the inner pole, for example, by at least a factor of 1.5 and preferably 2. The closed shape of the magnetron lessens electron loss in the high density plasma adjacent the target. The unbalanced magnetic field results in magnetic field lines projecting far from the stronger outer pole towards wafer. The projecting field lines both support a more extensive plasma and guide the metal ions towards the wafer.
Reasonable levels of sputtering uniformity are achieved in SIP sputtering by rotating the small magnetron about the center of the target and by shaping the magnetron to favor the outer portions of the target. Preferably, the outer pole of the unbalanced magnetron has a generally triangular shape with a triangular inner aperture in which is disposed the inner pole. Apex angles for the most acute corner are typically around 20 to 35°. The acute apex of the triangular pole overlies or is close to the center of rotation. The base of the triangular pole is close to the outer periphery and may be curved to follow the target circumference.
Although the rotating triangular magnetron provides reasonably adequate uniformity, uniformity for thin barrier layers in high-aspect ratio holes is a complex requirement, as has been partially discussed with reference to FIG. 3. Sidewall coverage needs to be relatively high, and it needs to be uniform across the large wafer. Furthermore, the sidewall coverage on one sidewall should not differ significantly from sidewall coverage on the opposed sidewall. The projecting magnetic field from the unbalanced triangular magnetron is very non-uniform in the radial direction, and its non-uniformity cannot be compensated by only circumferential scanning. The triangular design by itself is constrained for improving the many factors of uniformity and deep hole coating. Various types of auxiliary magnets have been proposed to compensate the inherent non-uniform magnetic field in a triangular magnetron, but these designs suffer their own deficiencies. Even a circular magnetron produces a magnetic field varying across its radius and its extensions.
Another problem with circumferentially scanned magnetrons is the typically non-uniform erosion in the radial direction. This problem arises even when the magnetron has a rather large size, such as the kidney-shaped magnetron. A typical erosion pattern 52 below an initial planar target surface 54 for a triangular SIP magnetron is illustrated in FIG. 4 for a magnetron having a target layer of the sputtering material bonded along an interface 56 to a backing plate of a different material. Distinctive annular trough-shaped erosion paths develop. It is difficult to achieve high utilization of the target center with only circumferential scanning of the small magnetron favored for SIP sputtering. Although the non-uniform erosion is reduced with the use of large kidney-shaped magnetron, it still occurs to a significant degree. The lifetime of the target is determined by the first exposure of the backing plate. Further sputtering would contaminate the wafer with the non-desired material of the backing plate, and the target must be discarded or at least refurbished with a new target layer. However, overall target utilization is poor, about 50% in the illustrated example. When an integral target is used without a distinct backing plate, as is typical for aluminum sputtering, the considerations are somewhat different, but poor target utilization resulting from erosion tracks is still a problem. It is greatly desired to achieve more uniform sputtering to avoid excessive expense in replacing targets.
A problem arises with magnetron sputtering being used for a variety of applications with differing requirements on the shape and intensity of the magnetic field. While satisfactory sputtering reactors have been designed for most of these applications, often the reactors and their magnetrons have substantially different designs. The increasing number of different types of reactors and magnetrons imposes economic and inventory penalties in designing, distributing, and maintaining so many different types of reactors. It is thus desired to obtain universal sputter reactor and magnetron designs in which small changes in the design or changed operational parameters allows the same design to be used in disparate applications.
Various suggestions have been made to scan a magnetron in both circumferential and radial directions about a circular target, typically in an epicyclic pattern of a primary rotation about the target center and a secondary rotation about the end of the arm of the primary rotation. See for example, U.S. Pat. No. 4,714,536 to Freeman et al. and U.S. Pat. No. 5,126,029 to Tomer et al. The Freeman design seems more practical, but it suffers from an inability to rotate the magnetron over the target center, and it is prone to excessive vibration. The Tomer design allows for center scanning, but its stationary internally toothed circumferential gear is unwieldy. The Tomer design is directed to smoothing non-uniform erosion tracks produced by a larger magnetron.